Data audit (2026-10-30)
All SHA-256 match data/README.md and bundle.json.
| file | data rows | cols | range | NA | dup dates | gaps |
|---|---|---|---|---|---|---|
| vintages/..._2024-01-31.csv | 307 | 2 | 1947-04-01..2023-10-01 | 0 | 0 | none |
| vintages/..._2024-04-30.csv | 308 | 2 | ..2024-01-01 | 0 | 0 | none |
| vintages/..._2024-07-31.csv | 309 | 2 | ..2024-04-01 | 0 | 0 | none |
| vintages/..._2024-10-31.csv | 310 | 2 | ..2024-07-01 | 0 | 0 | none |
| vintages/..._2025-01-31.csv | 311 | 2 | ..2024-10-01 | 0 | 0 | none |
| vintages/..._2025-04-30.csv | 312 | 2 | ..2025-01-01 | 0 | 0 | none |
| vintages/..._2025-07-31.csv | 313 | 2 | ..2025-04-01 | 0 | 0 | none |
| vintages/..._2025-12-31.csv | 314 | 2 | ..2025-07-01 | 0 | 0 | none |
| current_vintage.csv | 318 | 2 | 1947-04-01..2026-07-01 | 0 | 0 | none |
| first_release.csv | 13 | 3 | quarters 2023-07-01..2026-07-01 | 0 | 0 | none |
Units: % change from preceding quarter, SAAR (percentage points). SHA-256 values: see README (verified with shasum -a 256).
Anomalies:
- COVID-19 quarters 2020Q2 (≈ −28) and 2020Q3 (≈ +35) are extreme outliers in all vintages; they would dominate OLS AR estimates. Handling: exclude regression observations whose dependent variable or lag falls in 2020Q2–2020Q3 (rows t = 2020Q2, 2020Q3, 2020Q4). Decision made before computing any forecast.
- Origin spacing irregular: 8th origin is 2025-12-31 (advance 2025Q3 delayed by shutdown), per README. No handling needed; targets 2025Q4 and 2026Q1 have first releases.
- Revisions between vintages are visible (e.g. 2023Q4: 3.3 → 3.4 → 3.2; 2025Q2: 3.0 → 3.8) — this is the real-time feature, handled by using each origin's own vintage.
- All 16 target quarters (2024Q1..2026Q1) are present in first_release.csv. first_release_vintage_date of each h=1 target is after its origin (checked in analysis).
Availability: current_vintage.csv violates real-time availability for all origins → used only for revision statistic. first_release.csv values are post-origin by construction → used only as actuals.